Least squares adjustments and controlling control

[DanNarsavage]

I apologize in advance for the tome I'm about to unleash upon you all.  Once you've had your fill of my introduction of the problem, you can skip to the bold bits for my questions.

For decades, our surveyor has maintained a dataset of PLS line measurements and corner coordinates in WinGMM.  He likes it, and its LSA results have done well to improve our map (our populated areas have been nailed down pretty well but out in the hinterlands where our knowledge of section corners is lackluster, it's consistently pulled things from hundreds of feet away from aerials to within a few feet of aerials).  These LSAs are based largely on bearings & distances of section lines that he finds on plats & records of survey (which are free but rarely publish coordinates) and refined using GPS observations from licensed contractors (which aren't in the budget anymore) where available.

While our parcels were maintained in coverages, my predecessor would obtain section corner coordinates from the results of our surveyor's WinGMM LSAs and (I think) simply rubbersheet our parcels & PLS layers to fit the new corner coordinates.  When we migrated to a parcel fabric, I hoped to employ that same process (using those GMM-derived coordinates as control points) but to make the "rubbersheeting" both quicker and more rigorous by running a LSA in the fabric.  But I've run into some serious problems with fabric LSAs.  First is that WinGMM is eventually gonna have to go away (the surveyor's retirement or when GMM no longer runs on our OS, whichever's earlier) so we need to find a new solution for "controlling" our control.  My first thought on that was to use a "survey dataset" to manage control points (I think that was ESRI's intent for that data structure, yes?).  But that technology's been abandoned, so instead we've been trying to work on creating a separate parcel fabric that only contains PLS section lines (possibly down to QQs if needs be) that will replace GMM.  That plan has also hit a snag, and it's because of the second issue:

After two or three years of working with parcel fabrics, I have yet to run a least squares adjustment in the parcel fabric that I feel completely comfortable with.  Recently, I've taken the steps of writing a script that makes parcel fabric "import source" feature classes from WinGMM flat files so that I can compare the results of LSAs run in GMM and in a parcel fabric using the EXACT same measurement data.  The results of said comparison have been less than ideal, and I have a couple ideas as to why:

1) I'm less than ideal as a person to be running a LSA.  Most of the math going on inside an LSA is beyond my comfort zone so I don't really know how to set those tolerances.  Our surveyor is more comfortable with the math (kinda his gig, y'know?) but the settings are so wildly different from what he's used to seeing that he's thrown up his hands trying to figure it out.  Consequently I just jack up the tolerances until the LSA doesn't fail, then ratchet them back down until it does fail, then nudge them back up a little.

So my primary question is this: Has anyone had any better luck than I have with LSAs in similar situations (limited GPS coordinates & older measurements)?  Are there tricks that I'm missing?

Continuing with a bit of a wish list (I'll also post this to ideas.esri.com) . . .

2) It appears to me that the fabric's LSA has been oversimplified (relative to GMM's anyway)--it relies heavily on tight GPS control, assumes control points are positioned perfectly, and assumes all bearings & distances are modern.  The help doc (link) says that (regardless of the years shown alongside the ratings) the fabric treats lines with "1" accuracy as having an uncertainty of 5ppm, which even *I* can tell is ludicrous.  Modern legal descriptions often do close that tightly, but they're not actual field measurements--they're the results of calculations based on field observations (usually via GPS rather than chains & theodolites).  I learned in high school chemistry that you can't report more significance than you measure in the first place.  Our office requires all new legals to close to 1:5000 (200ppm), and I think this is industry standard, so the very best that we can do is to stretch the truth & call modern legal descriptions a "4."  But the vast majority of our county is described only by the original survey--performed over a century ago, most of it in great haste by the lowest bidder, and often with an order of magnitude more uncertainty than the "7" accuracy range's cutoff.  The fabric won't even attempt a LSA on an area with that much uncertainty, so we're up a creek.  And yet GMM gets the job done adequately with limited modern information when run by someone who understands what's going on under the hood.


So also, is there work being done to allow fabric LSAs to be more robust? Specifically, I'd like to see these improvements:
- reduce reliance upon new GPS observations in favor of bearing/distance measurements
- allow finer control over accuracy (GMM allows users to specify ppm values rather than limiting us to six confidence ratings)
- make use of older, less accurate measurements in LSAs
- accommodate basis-of-bearing bearing information so absolute bearings (rather than just internal angles) can be considered in LSAs
- employ control point uncertainty estimates to allow control points to wander within a predetermined envelope during LSAs

Thanks,
Dan

[ChristineLeslie]

Hi Dan

Just FYI, you can edit the ppm and standard deviation values in the Accuracy table. YOu are limited to 7 categories, but you are not limited to the values you can use. If you edit the ppm and std deviation values, the LSA will use them instead of the defaults.
Use the Make Parcel Fabric Table View GP tool to view and edit the Accuracy table

Christine

[ChristineLeslie]

I feel I should also add that you are really only limited to 6 categories as the 7th category is an "exclude" category - lines/parcels with the 7th category do not influence the outcome of the adjustment but will still move with connected points.
Also when editing ppm values (which should only be done by those people who thoroughly understand these numbers and how they are used), they still should reflect the accuracy level and be in an order of highest to lowest - ie ppm/std dev values for accuracy 3 should not be higher than those for accuracy 1.

Christine

[TimHodson]

Hi Dan,

Further to our e-mail correspondence, wanted to share the info (so far) on the forum. Look forward to talking with you about this further in the near future. Again, thanks for the feedback! As I mentioned in the e-mail, this does not answer all your questions, but we can take it further once we�??ve looked at your data. You are correct in that the fabric adjustment does not have the same functionality as WinGMM (such as a weighted least squares) We�??d be interested to see the data and results of the work that you and Jack have done in your comparisons between WinGMM and fabric adjustment.

While it is possible to express the closure of a legal description (say 1:5000) as a ppm, this is not the same as the standard deviation and ppm as found in the fabric�??s accuracy table described by Christine.

The legal description provides an expression of intent for what the surveyor placed/found in the field. It is even possible for the legal description �??measurement�?� values to be created before the surveyor physically places the new monuments, and before any direct field measurement is ever made. The mathematical closure of the legal description (regardless of age) may often be less than a 1/100th of a foot, however that does not mean the monuments as directly measured and placed in the field are �??accurate to a hundredth of a foot.�?� How well the actual positions of the monuments in the field match with the legal description�??s indirect measurements is a function of the equipment used by the surveyor to directly measure them. We would absolutely prefer to use the direct survey measurements, but since we do not have those original field observations, the legal record values are the next best thing. The other information that we have is the date of survey. From this we can infer the survey equipment that was likely used. The standard deviations and ppm from the accuracy table of the fabric is used to inform the fabric adjustment how well we expect the record values on each line to correlate with the actual �??true�?� value between the physical positions on the ground. (�??True�?� is in quotes here because it�??s always a theoretical value; a measurement by definition, always has error, whether it�??s indirect or direct.)

-Tim

[DanNarsavage]

FINALLY getting to read the replies.  Thanks both of you.  Regarding tweaking the PPM values, there's the first instance of what's sure to be many inadequacies of the user.  🙂  That may be a crucial improvement to our process, but we'll have to see.  I'm working on getting a bunch of data together before we chat.  You should have that soon.  Thanks again!

[deleted-user-V04hurjTRt9S]

I have recently converted some data into the LGIM but am having no luck with running LSA.  The county has a group of control points that were sent after the compilation of the data was complete.  Their hope was to load the new control points into the fabric and then run LSA to adjust the parcel and sub poly data to the new points.  I believe that I have followed the Help Documentation exactly, but I am unable to get the LSA to run without failing.  No matter what I try, it always fails with the following messages:


ERROR: Unable to complete adjustment. There are too many parcels, or not enough control points. The upper limit of parcels that can be adjusted is 5000. Add more control or reduce the number of parce
ERROR: Unable to complete adjustment. You are trying to adjust too many parcels. The upper limit of parcels that can be adjusted is 5000. Add more control or reduce the number of parcels being adjusted.

I am only attempting to adjust 10 parcels and I have even tried manually adding several control points matching existing parcel points.  I then attempted to perform the LSA in a test dataset that we have been using for testing LSA for the past few years and it no longer works in that dataset either.  I guess my question then is this:  Is anyone successfully running LSA in Arc10 SP4?  If so please provide any tips or tricks that you may have.

[ChrisBuscaglia]

This is a known issue, what you are experiencing is an incorrect error message.  This is meant to report that you are attempting to adjust parcels/lines with accuracy 7.

http://help.arcgis.com/en/arcgisdesktop/10.0/help/#/About_accuracy_in_the_parcel_fabric/001t00000145...